Pierre Gançarski

A global averaging method for dynamic time warping, with applications to clustering

Mining sequential data is an old topic that has been revived in the last decade, due to the increasing availability of sequential datasets. Most works in this field are centred on the definition and use of a distance (or, at least, a similarity measure) between sequences of elements. A measure called dynamic time warping (DTW) seems to be currently the most relevant for a large panel of applications. This article is about the use of DTW in data mining algorithms, and focuses on the computation of an average of a set of sequences. Averaging is an essential tool for the analysis of data. For example, the K-means clustering algorithm repeatedly computes such an average, and needs to provide a description of the clusters it forms. Averaging is here a crucial step, which must be sound in order to make algorithms work accurately. When dealing with sequences, especially when sequences are compared with DTW, averaging is not a trivial task. Starting with existing techniques developed around DTW, the article suggests an analysis framework to classify averaging techniques. It then proceeds to study the two major questions lifted by the framework. First, we develop a global technique for averaging a set of sequences. This technique is original in that it avoids using iterative pairwise averaging. It is thus insensitive to ordering effects. Second, we describe a new strategy to reduce the length of the resulting average sequence. This has a favourable impact on performance, but also on the relevance of the results. Both aspects are evaluated on standard datasets, and the evaluation shows that they compare favourably with existing methods. The article ends by describing the use of averaging in clustering. The last section also introduces a new application domain, namely the analysis of satellite image time series, where data mining techniques provide an original approach.

Collaborative clustering with background knowledge

The aim of collaborative clustering is to make different clustering methods collaborate, in order to reach at an agreement on the partitioning of a common dataset. As different clustering methods can produce different partitioning of the same dataset, finding a consensual clustering from these results is often a hard task. The collaboration aims to make the methods agree on the partitioning through a refinement of their results. This process tends to make the results more similar. In this paper, after the introduction of the collaboration process, we present different ways to integrate background knowledge into it. Indeed, in recent years, the integration of background knowledge in clustering algorithms has been the subject of a lot of interest. This integration often leads to an improvement of the quality of the results. We discuss how such integration in the collaborative process is beneficial and we present experiments in which background knowledge is used to guide collaboration.

Extraction of complex patterns from multiresolution remote sensing images: A hierarchical top-down methodology

The extraction of urban patterns from very high spatial resolution (VHSR) optical images presents several challenges related to the size, the accuracy and the complexity of the considered data. Based on the availability of several optical images of a same scene at various resolutions (medium, high, and very high spatial resolution), a hierarchical approach is proposed to progressively extract segments of interest from the lowest to the highest resolution data, and then finally determine urban patterns from VHSR images. This approach, inspired by the principle of photo-interpretation, has for purpose to use as much as possible the users skills while minimising his/her interaction. In order to do so, at each resolution, an interactive segmentation of one sample region is required for each semantic class of the image. Then, the users behaviour is automatically reproduced in the remainder of the image. This process is mainly based on tree-cuts in binary partition trees. Since it strongly relies on user-defined segmentation examples, it can involve only low level-spatial and radiometric-criteria, then enabling fast computation of comprehensive results. Experiments performed on urban images datasets provide satisfactory results which may be further used for classification purpose.

Knowledge-based region labeling for remote sensing image interpretation

The increasing availability of High Spatial Resolution (HSR) satellite images is an opportunity to characterize and identify urban objects. Thus, the augmentation of the precision led to a need of new image analysis methods using region-based (or object-based) approaches. In this field, an important challenge is the use of domain knowledge for automatic urban objects identification, and a major issue is the formalization and exploitation of this knowledge. In this paper, we present the building steps of a knowledge-base of urban objects allowing to perform the interpretation of HSR images in order to help urban planners to automatically map the territory. The knowledge-base is used to assign segmented regions (i.e. extracted from the images) into semantic objects (i.e. concepts of the knowledge-base). A matching process between the regions and the concepts of the knowledge-base is proposed, allowing to bridge the semantic gap between the images content and the interpretation. The method is validated on Quickbird images of the urban areas of Strasbourg and Marseille (France). The results highlight the capacity of the method to automatically identify urban objects using the domain knowledge.

Ontology-Based Object Recognition for Remote Sensing Image Interpretation

The multiplication of very high resolution (spatial or spectral) remote sensing images appears to be an opportunity to identify objects in urban and periurban areas. The classification methods applied in the object-oriented image analysis approach could be based on the use of domain knowledge. A major issue in these approaches is domain knowledge formalization and exploitation. In this paper, we propose a recognition method based on an ontology which has been developed by experts of the domain. In order to give objects a semantic meaning, we have developed a matching process between an object and the concepts of the ontology. Experiments are made on a Quickbird image. The quality of the results shows the effectiveness of the proposed method.

Summarizing a set of time series by averaging: From Steiner sequence to compact multiple alignment

Summarizing a set of sequences is an old topic that has been revived in the last decade, due to the increasing availability of sequential datasets. The definition of a consensus object is on the center of data analysis issues, since it crystallizes the underlying organization of the data. Dynamic Time Warping (DTW) is currently the most relevant similarity measure between sequences for a large panel of applications, since it makes it possible to capture temporal distortions. In this context, averaging a set of sequences is not a trivial task, since the average sequence has to be consistent with this similarity measure. The Steiner theory and several works in computational biology have pointed out the connection between multiple alignments and average sequences. Taking inspiration from these works, we introduce the notion of compact multiple alignment, which allows us to link these theories to the problem of summarizing under time warping. Having defined the link between the multiple alignment and the average sequence, the second part of this article focuses on the scan of the space of compact multiple alignments in order to provide an average sequence of a set of sequences. We propose to use a genetic algorithm based on a specific representation of the genotype inspired by genes. This representation of the genotype makes it possible to consistently paint the fitness landscape. Experiments carried out on standard datasets show that the proposed approach outperforms existing methods.

Multiresolution Remote Sensing Image Clustering

With the multiplication of satellite images with complementary spatial and spectral resolution, a major issue in the classification process is the simultaneous use of several images. In this context, the objective of this letter is to propose a new method which uses information contained in both spatial resolutions. The main idea is that on one hand, the semantic level associated with an image depends on its spatial resolution, and on the other hand, information given by these images is complementary. The goal of this multiresolution image method is to automatically build a classification using knowledge extracted from both images, by unsupervised way and without preprocessing image fusion. The method is tested by using a Quickbird (2.8 m) and a SPOT-4 (20 m) image on the urban area of Strasbourg (France). The experiments have shown that the results are better than a classical unsupervised classification on each image and comparable to a supervised region-based classification on the high-spatial-resolution image.

Combat or surveillance? Evaluation of the heterogeneous inflammatory breast cancer microenvironment

Evaluation of specific lymphocyte subsets is important in understanding the microenvironment in cancer and holds promise as a prognostic parameter in invasive breast cancer. To address this, we used digital image analysis to integrate cell abundance, distance metrics, neighbourhood relationships and sample heterogeneity into comprehensive assessment of immune infiltrates. Lymphocyte and macrophage subpopulations were detected by chromogenic duplex immunohistochemistry for CD3/perforin and CD68/CD163 in samples of invasive breast cancer. The analysis workflow combined commercial and open‐source software modules. We confirmed the accuracy of automated detection of cells with lymphoid morphology [concordance correlation coefficient (CCC), 0.92 for CD3+‐T lymphocytes], whereas variable morphology limited automated classification of macrophages as distinct cellular objects (CCC, 0.43 for object‐based detection; 0.79 for pixel‐based area analysis). Using a supervised learning algorithm that clustered image areas according to lymphocyte abundance, grouping behaviour and distance to tumour cells, we identified recurrent infiltration patterns reflecting different grades of direct interaction between tumour and immune effector cells. The approach provided comprehensive visual and statistical assessment of the inflammatory tumour microenvironment and allowed quantitative estimation of heterogeneous immune cell distribution. Cases with dense lymphocytic infiltrates (8/33) contained up to 65% of areas in which observed distances between tumour and immune cells suggested a low chance of direct contact, indicating the presence of regions where tumour cells might be protected from immune attack. In contrast, cases with moderate (11/33) or low (14/33) lymphocyte density occasionally comprised areas of focally intense interaction, likely not to be captured by conventional scores. Our approach improves the conventional evaluation of immune cell density scores by translating objective distance metrics into reproducible, largely observer‐independent interaction patterns.

MACLAW: A modular approach for clustering with local attribute weighting

This paper presents a new process for modular clustering of complex data, like remote sensing images. This method performs feature weighting in a wrapper approach. The proposed method is a modular clustering method that combines several extractors, which are local specialists, each one extracting one cluster only and using different feature weights. A new clustering quality criterion, adapted to independent clusters, is defined. The weight learning is performed through a cooperative coevolution algorithm, where each species represents one of the clusters to be extracted. A set of extracted clusters forms a partial soft clustering but can be transformed in a classic hard clustering. Some tests, on datasets from the UCI repository and on hyperspectral remote sensing image, have been performed and show the validity of the approach.

Multi-resolution region-based clustering for urban analysis

In the domain of urban planning and management, it may be necessary to map the territory at different scales, each corresponding to a semantic level. Three semantic levels are identified: (1) the object level, for mapping urban elements (buildings, etc.), (2) the block level, for mapping homogeneous patterns of urban elements, and (3) the area level, for mapping urban fabrics defined as sets of homogeneous patterns. Some of these levels are directly linked to specific satellite images presenting ad hoc resolutions (namely, medium spatial resolution (MSR) images for the area level and high spatial resolution (HSR) images for the object level); in such cases, a straightforward mapping can be performed by clustering the data. Conversely, classical clustering techniques do not enable the intermediate semantic level to be extracted directly. The purpose of this article is to propose a methodology enabling a clustering at this level to be generated. The proposed approach is, in particular, based on the segmentation and unsupervised, region-based and joined clustering of two images representing a same scene at MSR and HSR. The method has been applied to different and heterogeneous datasets composed of HSR images at 2.5 m and MSR images at 10 m and 20 m. Qualitative validations by an expert, and quantitative ones by comparison to other existing methods, tend to emphasize the soundness and efficiency of this methodology, thus justifying further developments.